se-280 dr. mark l. hornick numerical integration

15
SE-280 Dr. Mark L. Hornick Numerical Integration

Upload: isabella-wiggins

Post on 26-Dec-2015

220 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

Numerical Integration

Page 2: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

2

In the PSP, definite integrals of the t-distribution are used to calculate the significance of a correlation and the prediction interval of an estimate.

Requirement: Integrate an arbitrary f(x)

from a to b

( ) ( )

( ) ( ) ( )b

a

F x f x dx

f x dx F b F a

The problem is that there is no (simple) closed-form solution for the integral of the t-distribution function.

Page 3: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

3

When is Numerical Integration needed?

Analytic solution F(x) is not always practical

2

???xF x e dx

Page 4: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

4

Numerical Integration Approach

Fit polynomial (or something else) to f(x) All at once In discrete segments Polynomial degree can vary

Integrate resulting polynomial(s) Using well-known formulas

Page 5: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

5

0

0.2

0.4

0.6

0.8

1

1.2

0 1 2 3

f(x)

Integration Example

2xf x e

Page 6: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

6

0

0.2

0.4

0.6

0.8

1

1.2

0 1 2 3

f(x)

Zeroth-Order Fit

h

Page 7: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

7

Zeroth-Order Fit

h h

f0

f2

f1

0h f 1h f

Page 8: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

8

0

0.2

0.4

0.6

0.8

1

1.2

0 1 2 3

f(x)

Zeroth-Order Fit

0

0 1 1

nx

n

x

f x dx h f f f

hNote: Sometimes “w” is used instead of “h” for step width.

Page 9: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

9

First-Order Fit (trapezoidal rule)

h h

f0

f2

f1

0 1

2

f fh

1 2

2

f fh

Page 10: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

10

0

0.2

0.4

0.6

0.8

1

1.2

0 1 2 3

f(x)

First-Order Fit (contd.)

0

0 1 12 22

nx

n n

x

hf x dx f f f f

Page 11: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

11

Second-Order Fit

h

f0

f2

f1

h

Parabolic curve

Page 12: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

12

Simpson’s Rule

Simpson’s 1/3 rule: parabolic segments

0

0 1 2

2 1

4 2

2 43

nx

n n nx

f f fhf x dx

f f f

CoefficientsFirst and last terms: 1Odd terms: 4Even terms: 2n must be even

Page 13: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

13

In all these methods, we must choose an appropriate step size.

A small step size generally gives a better fit, but takes longer to calculate and may increase round-off error.

A large step size is usually less accurate, but faster to compute.

Page 14: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

14

Often the best choice is to iterate to the "right" step size.

Choose # ofsegments (n1)

Calculate integral(save as pa)

Calc new segments(nj = 2*nj-1)

|pb - pa| < eCalculate integral

(save as pb)

Set pa = pb

Done (answer = pb)"e" is the desired result precision

"h" is derived from "n", the number of

segments

Yes

No

Page 15: SE-280 Dr. Mark L. Hornick Numerical Integration

SE-280Dr. Mark L. Hornick

15

Pro

bab

ilit

y d

ensi

ty

Distributions are important statistical functions that we often need to integrate numerically, since no closed-form solution exists.

2

21

2

x

F x e

Normal Distribution: The probability density function for a large sample size

Its integral represents a cumulative probability over some range (more on that in a later).